From 2 December 2026 it will be prohibited in the Union to place on the market, put into service or use artificial intelligence systems that generate or manipulate realistic images, video and audio of the intimate parts of an identifiable person, or depicting that person engaged in sexually explicit activities, without his or her consent. The prohibition was inserted into Article 5 of Regulation (EU) 2024/1689 (AI Act) by Regulation (EU) 2026/1744 of 8 July 2026, the Digital Omnibus on AI, which entered into force on 27 July 2026 (EUR-Lex; European Commission). Recital 11 is explicit: “Non-consensual intimate material constitutes sexual violence and abuse against individuals, in particular women.” The new rule sits alongside Directive (EU) 2024/1385 on combating violence against women and domestic violence, not yet transposed in Italy, and Article 612-quater of the Italian Criminal Code.
The new AI Act prohibition: providers, deployers and safety measures
The new point (ba) of Article 5(1) prohibits “the placing on the market, the putting into service or the use of an AI system that generates or manipulates realistic images, videos, audio or similar material of an identifiable natural person’s intimate parts, or of an identifiable natural person engaged in sexually explicit activities” in the absence of that person’s “freely-given, specific, informed, unambiguous and explicit consent” (EUR-Lex). Point (bb) extends the prohibition to child sexual abuse material.
As Nicola Fabiano points out, the Italian text of the provision departs from the English text on two elements of the prohibition (NicFab): the English version quoted above requires consent that is also “explicit” and refers the prohibition to an “identifiable natural person” (EUR-Lex), whereas the Italian enacting terms omit the requirement of explicitness, referring to consent that is only “liberamente prestato, specifico, informato e inequivocabile” (freely given, specific, informed and unambiguous), and speak of a “persona fisica riconoscibile” (a recognisable natural person); the Italian recital 12, however, requires consent that is “inequivocabile ed esplicito” (unambiguous and explicit) and, like recital 14, refers to “identificabili” (identifiable) persons, in line with the English text. Since the language versions are equally authentic, we consider that providers should adopt the standard of explicit consent and the broader scope of identifiability. Fabiano adds that, where the material constitutes data concerning a person’s sex life, explicit consent is in any event required by Article 9(2)(a) of Regulation (EU) 2016/679 (GDPR).
Paragraph 1a distinguishes the two roles. For providers, placing on the market and putting into service are prohibited only if the generation is the intended purpose of the system, or if it is a reasonably foreseeable and reproducible outcome without significant technical modification and the system does not have reasonable and adequate safety measures to prevent it and to correct misuse. For deployers, use is prohibited only if the system is used for the purpose of generating or manipulating such material. Paragraph 1b excludes manipulations that do not increase the exposure of intimate parts or alter the nature of the activities depicted. The prohibition targets the purpose of the system and the absence of safeguards, not the technical capability to generate images. Recital 12 lists among the possible measures data cleaning, refusal training and content filtering.
The new provisions apply from 2 December 2026 by virtue of the new Article 113, third paragraph, point (a) (EUR-Lex). Under Article 99(3) of the AI Act, infringements are subject to administrative fines of up to EUR 35 000 000 or, for undertakings, up to 7 % of total worldwide annual turnover, whichever is higher (European Commission). In Italy supervision lies with the National Cybersecurity Agency (ACN) under Article 20(1)(b) of Law No 132 of 23 September 2025 (Official Gazette (Gazzetta Ufficiale)), and Legislative Decree No 179 of 7 October 2026, in force from 23 October 2026, designates it as market surveillance authority (Article 5); Article 23(1) of the same Decree provides that, “unless the act constitutes a criminal offence” (our translation), non-compliance with the prohibition of the practices referred to in Article 5 of the Regulation is subject to the same maximum amounts (Official Gazette (Gazzetta Ufficiale)). The Decree does not refer to the Digital Omnibus on AI, but in our view the reference to Article 5 also covers the new points.
Directive (EU) 2024/1385: harmonised offences, removal orders and transposition
Directive (EU) 2024/1385 operates on a different level: its Article 5(1) requires Member States to make three forms of intentional conduct punishable as criminal offences: making accessible to the public, by means of information and communication technologies, material depicting sexually explicit activities or the intimate parts of a person without that person’s consent (point (a)); “producing, manipulating or altering and subsequently making accessible to the public, by means of ICT, images, videos or similar material making it appear as though a person is engaged in sexually explicit activities”, without consent (point (b)); threatening such conduct in order to coerce the person into a certain behaviour (point (c)). For points (a) and (b) it is sufficient that the conduct is likely to cause serious harm; recital 19 brings within point (b) deepfakes depicting a person performing sexual acts, including where produced with artificial intelligence (EUR-Lex).
Article 23 requires the material referred to in points (a) and (b) to be promptly removed or access to it disabled, without prejudice to Regulation (EU) 2022/2065 (Digital Services Act), including through binding orders of the competent authorities which “shall be addressed to hosting service providers”; the orders are revoked if the criminal proceedings end without a finding that an offence was committed. The transposition deadline is 14 June 2027 (Article 49). The Directive is included in Annex A to Law No 91 of 13 June 2025, the 2024 European Delegation Law (Official Gazette (Gazzetta Ufficiale)); as at 12 October 2026 the EUR-Lex database shows no transposition measures notified by Italy (EUR-Lex). Until transposition, the removal orders under Article 23 have no national implementing rules.
The Italian criminal-law response: Article 612-quater and corporate liability
Article 612-quater, introduced by Article 26(1)(c) of Law No 132/2025 and in force since 10 October 2025, punishes with imprisonment of one to five years “Anyone who causes unjust harm to a person by transferring, publishing or otherwise disseminating, without that person’s consent, images, videos or voices falsified or altered through the use of artificial intelligence systems and capable of misleading as to their genuineness” (Official Gazette (Gazzetta Ufficiale)). The offence is not limited to sexual content and requires harm actually caused. In our view the provision covers a space left open by Article 612-ter, which concerns sexually explicit images or videos made or stolen and “intended to remain private” (Normattiva): a requirement that can hardly be applied to synthetic content.
Legislative Decree No 160 of 9 September 2026, in force since 30 September 2026, adds two elements (Official Gazette (Gazzetta Ufficiale)). Article 15 inserts into Legislative Decree No 231 of 8 June 2001 a new Article 25-vicies, which for the offence under Article 612-quater provides for a financial penalty on the entity of two hundred to seven hundred units and for the disqualification sanctions of Article 9(2)(b), (c), (d) and (e). Article 14 amends Article 104(1)(e-bis) of the implementing provisions of the Code of Criminal Procedure, extending preventive seizure to online content “generated also with artificial intelligence systems”.
Three levels of protection: the tool, the conduct, the intermediary
The AI Act acts upstream, on the tool, with administrative fines on providers and deployers; criminal law acts on the person who disseminates the content; removal orders act on intermediaries, currently within the framework of preventive seizure and, in future, under Article 23. On coordination, recital 15 of the Digital Omnibus on AI specifies that the prohibitions “do not preclude prosecution” and that, where the same conduct is also sanctioned under criminal law, including criminal law implementing Directive (EU) 2024/1385, Member States must ensure respect for the ne bis in idem principle; recital 16 preserves national remedies protecting image, privacy and dignity. The three levels do not replace one another and operate on different addressees and through different authorities. On the allocation of responsibilities along the value chain we refer to our analysis of who answers when the system acts; on the scale of the phenomenon, to the figures already examined on deepfakes and technology-facilitated gender-based violence and on online violence against girls.
Critical aspects: timing, addressees and thresholds of criminal liability
The first aspect is timing. The AI Act prohibition will apply from 2 December 2026, more than six months before the transposition deadline of the Directive, and until then harmonised offences and removal orders will depend on national implementation.
The second concerns the addressees: the definition of deployer excludes use “in the course of a personal non-professional activity” (Article 3, point (4)) and Article 2(10) excludes the obligations of deployers who are natural persons using AI systems in a purely personal activity (European Commission); the Digital Omnibus on AI has not amended these provisions. The private individual who generates a fake intimate image for personal use remains, in our view, outside the prohibition on use. In criminal law, transfer, publication or dissemination is required: mere generation constitutes neither the offence under Article 612-quater nor those under points (a) and (b) of Article 5 of the Directive.
The third concerns the alignment between the Italian provision and the Directive. Article 612-quater requires actual harm, whereas the Directive is satisfied with conduct likely to cause serious harm; it is limited to content obtained with AI, whereas the Directive covers any manipulation; it does not cover the threat of dissemination under point (c). Moreover, Article 612-quater does not provide for the aggravating circumstance that Law No 181 of 2 December 2025 inserted into Article 612-ter for acts committed “as an act of hatred or of discrimination or of oppression or as an act of control or possession or domination because she is a woman” (Normattiva). In Union law too the notions do not coincide: point (ba) of the AI Act also covers intimate parts alone, whereas point (b) of the Directive covers only material simulating sexually explicit activities. In our view transposition is the opportunity to bring these differences into a coherent design.
Conclusions
From 2 December 2026 providers will not be able to place on the market systems intended to produce non-consensual intimate material, or that produce it foreseeably without adequate safeguards; those who disseminate the content are already liable under Article 612-quater and, since 30 September 2026, the entity may also be liable under Legislative Decree 231/2001. Directive (EU) 2024/1385 must be transposed by 14 June 2027: in our view the legislative decree should clarify the relationship with Article 612-quater, regulate removal orders also outside criminal seizure and cover the threat of dissemination. Those who develop or integrate generative models would do well to document the preventive measures indicated in recital 12 now.
Author: Valentina Grazia Sapuppo
